Connect Cursor, VS Code and Gemini CLI to Roma

Cursor, VS Code and Gemini CLI connect to Roma the way any MCP client does: point them at https://api.roma.app/mcp, sign in to Roma when they ask, and Roma's tools are there. Cursor and VS Code have a one-click install link on their pages at /agents/cursor and /agents/vs-code. Gemini CLI takes one command, or the message on /agents/gemini-cli. Agents in Roma explains what a connected agent is.

Cursor

Use the one-click link, Add Roma to Cursor, on /agents/cursor, and sign in when Cursor asks. Or add this to ~/.cursor/mcp.json, or to the .cursor/mcp.json beside the code you are working in:

{
  "mcpServers": {
    "roma": { "url": "https://api.roma.app/mcp" }
  }
}

VS Code

Use the install link, Add Roma to VS Code, on /agents/vs-code, or run this from a terminal:

code --add-mcp '{"name":"roma","type":"http","url":"https://api.roma.app/mcp"}'

Confirm the trust prompt, then sign in through the browser VS Code opens on the first connection. The same entry by hand, in .vscode/mcp.json:

{
  "servers": {
    "roma": { "type": "http", "url": "https://api.roma.app/mcp" }
  }
}

Gemini CLI

One command adds the server:

gemini mcp add --transport http roma https://api.roma.app/mcp

Then /mcp auth roma inside the CLI to sign in. The gemini command serves organizations on Gemini Code Assist Standard or Enterprise and paid Gemini API keys. An individual Google account uses Antigravity CLI instead, where npx add-mcp (below) writes Roma into its config. Gemini CLI also takes the connect message from its page: send it, and the agent sets itself up.

Every coding agent on the computer at once

One command writes Roma into every coding agent it finds on the computer, Claude Code, Codex, Cursor, VS Code, Gemini CLI and Windsurf among them:

npx add-mcp https://api.roma.app/mcp -n roma

Each agent then runs its own sign-in the first time it connects.

What the sign-in looks like

The first connection opens a browser on Roma's sign-in. You sign in with Google or Apple, then approve on a screen that names the app and says where it returns you, which for these agents reads "an app on this computer". Nothing is pasted and no key is copied. The sign-in lasts an hour at a time and the agent renews it on its own. A machine with no browser takes an API key as a header instead, and all three agents have a place for one: API keys and signing in from an agent.

To check any of them, ask "What's on my plate today in Roma?" The answer should use your own task titles.

What a coding agent does well with Roma

A coding agent works from a task rather than only reminding you of it. Noor files a task from her phone, "Fix the login timeout, see the chat for the trace". In Cursor she asks it to pick that task up. It reads the task's body and chat, sets the status to In progress, does the work in the body, continuing what is there, and leaves one note in the task's chat, under its own name, with the link to the pull request it opened, so she sees it on every device. Completing the task stays hers. What an agent can and cannot do in your account has the rules an agent is given, and The conversation on a task or note explains the chat it posts into.

An agent on your Mac can also be woken by Roma the moment a task changes, instead of checking on a schedule. The trigger holds the events until the agent asks for them, so a laptop that was asleep picks up where it left off: Triggers: wake your agent when something changes.

Questions people ask

A tool I expect is missing in Cursor or VS Code. Why?

The agent cached an older tool list when it connected. Refresh the Roma server in the agent's MCP settings and start a new conversation, and the current tools are there.

It worked, then stopped an hour later. What happened?

The sign-in lasts an hour and the agent renews it on its own with the refresh it was given. An agent that keeps nothing between sessions has to sign in again. For a machine that runs unattended, an API key does not expire and never needs a browser.

Do I need a separate Roma account for each agent?

No. One Roma account serves every agent. Each agent signs in on its own the first time it connects, and all of them see the same lists.

Can one agent see what another agent wrote?

Yes. Everything an agent writes lands in Roma itself, on your lists, so Cursor sees the task Claude added this morning and the note Gemini CLI saved yesterday. The Event Log shows which agent made each change.